imgaug_sequential_image_augmentation_pipeline_quickstart.py

1import numpy as np
2import imgaug as ia
3import imgaug.augmenters as iaa
4
5# load images
6# (assuming you have your images in a list or numpy array)
7# images = ... 
8# For this example, we create some dummy data:
9images = np.random.randint(0, 255, (16, 128, 128, 3), dtype=np.uint8)
10
11# Define the augmentation sequence
12seq = iaa.Sequential([
13    iaa.Fliplr(0.5), # horizontal flips
14    iaa.Crop(percent=(0, 0.1)), # random crops
15    # Small gaussian blur with random sigma between 0 and 0.5.
16    # But we only apply it to about 50% of all images.
17    iaa.Sometimes(
18        0.5,
19        iaa.GaussianBlur(sigma=(0, 0.5))
20    ),
21    # Strengthen or weaken the contrast in each image.
22    iaa.LinearContrast((0.75, 1.5)),
23    # Add gaussian noise.
24    # For 50% of all images, we sample the noise once per pixel.
25    # For the other 50% of all images, we sample the noise per pixel AND
26    # channel. This can change the color (not only brightness) of the pixels.
27    iaa.AdditiveGaussianNoise(loc=0, scale=(0.0, 0.05*255), per_channel=0.5),
28    # Make some images brighter and some darker.
29    # In 20% of all cases, we multiply the predictor variable of each pixel
30    # with a value between 0.8 and 1.2 and then add a value between -10 and 10.
31    iaa.Multiply((0.8, 1.2), per_channel=0.2),
32    # Apply affine transformations to each image.
33    # Scale/zoom them, translate/move them, rotate them and shear them.
34    iaa.Affine(
35        scale={"x": (0.8, 1.2), "y": (0.8, 1.2)},
36        translate_percent={"x": (-0.2, 0.2), "y": (-0.2, 0.2)},
37        rotate=(-25, 25),
38        shear=(-8, 8)
39    )
40], random_order=True) # apply augmenters in random order
41
42# Augment the images
43images_aug = seq(images=images)